Prediction for 1-minute Movements of TAIEX Futures Based on LSTM Model

• Main Authors: CHANG,CHIA-WEN, 張嘉文
• Other Authors: YAN, RU-FANG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/3a3zgw

碩士 === 國立臺北大學 === 統計學系 === 107 === In recent years, because of the advancement of technology and the advancement of hardware and software, artificial intelligence has become a hot topic and is used in various fields. Financial technology has created financial technology and high frequency trading. Investors also hope to increase the winning rate of investment in investment financial products through artificial intelligence, and the excess rate of return can be achieved regardless of fluctuations in the market. The main purpose of this study is to hope that through the TAIEX Futures, the transaction data for the period from 2012 to 2016.Based on different historical data, it is predicted 1 minute ups and downs in 2016, and used the Long Short-Term Memory (LSTM) for analysis.The data variables are open , high ,low ,close , and volume of futures trading. The length of the data is 5 minutes to predict the verification data for one minute. The trend is up and down, with an average forecasting power of about 44%.From the forecasting ability, the cumulative remuneration and the return on investment, the research method is higher than the buying and holding strategy. The length of the training period has a positive relationship with the forecasting ability. The 4-year long-term training data forecast is more than 1 The annual data is 0.78% higher. The average cumulative compensation is 145,800 yuan higher. The average return on investment is 0.68% higher.